vector-quantize-pytorch 1.22.2__tar.gz → 1.22.3__tar.gz

{vector_quantize_pytorch-1.22.2 → vector_quantize_pytorch-1.22.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vector-quantize-pytorch
-Version: 1.22.2
+Version: 1.22.3
 Summary: Vector Quantization - Pytorch
 Project-URL: Homepage, https://pypi.org/project/vector-quantize-pytorch/
 Project-URL: Repository, https://github.com/lucidrains/vector-quantizer-pytorch
@@ -324,7 +324,7 @@ This repository should also automatically synchronizing the codebooks in a multi
 
 <img src="./images/simvq.png" width="400px"></img>
 
-A <a href="https://arxiv.org/abs/2411.02038">new ICLR 2025 paper</a> proposes a scheme where the codebook is frozen, and the codes are implicitly generated through a linear projection. The authors claim this setup leads to less codebook collapse as well as easier convergence. I have found this to perform even better when paired with <a href="https://arxiv.org/abs/2410.06424">rotation trick</a> from Fifty et al., and expanding the linear projection to a small one layer MLP. You can experiment with it as so
+A <a href="https://arxiv.org/abs/2411.02038">new paper</a> proposes a scheme where the codebook is frozen, and the codes are implicitly generated through a linear projection. The authors claim this setup leads to less codebook collapse as well as easier convergence. I have found this to perform even better when paired with <a href="https://arxiv.org/abs/2410.06424">rotation trick</a> from Fifty et al., and expanding the linear projection to a small one layer MLP. You can experiment with it as so
 
 Update: hearing mixed results
 

{vector_quantize_pytorch-1.22.2 → vector_quantize_pytorch-1.22.3}/README.md

@@ -280,7 +280,7 @@ This repository should also automatically synchronizing the codebooks in a multi
 
 <img src="./images/simvq.png" width="400px"></img>
 
-A <a href="https://arxiv.org/abs/2411.02038">new ICLR 2025 paper</a> proposes a scheme where the codebook is frozen, and the codes are implicitly generated through a linear projection. The authors claim this setup leads to less codebook collapse as well as easier convergence. I have found this to perform even better when paired with <a href="https://arxiv.org/abs/2410.06424">rotation trick</a> from Fifty et al., and expanding the linear projection to a small one layer MLP. You can experiment with it as so
+A <a href="https://arxiv.org/abs/2411.02038">new paper</a> proposes a scheme where the codebook is frozen, and the codes are implicitly generated through a linear projection. The authors claim this setup leads to less codebook collapse as well as easier convergence. I have found this to perform even better when paired with <a href="https://arxiv.org/abs/2410.06424">rotation trick</a> from Fifty et al., and expanding the linear projection to a small one layer MLP. You can experiment with it as so
 
 Update: hearing mixed results
 

{vector_quantize_pytorch-1.22.2 → vector_quantize_pytorch-1.22.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "vector-quantize-pytorch"
-version = "1.22.2"
+version = "1.22.3"
 description = "Vector Quantization - Pytorch"
 authors = [
     { name = "Phil Wang", email = "lucidrains@gmail.com" }

{vector_quantize_pytorch-1.22.2 → vector_quantize_pytorch-1.22.3}/vector_quantize_pytorch/residual_vq.py

@@ -85,6 +85,8 @@ class MLP(Module):
     ):
         one_headed = codes.ndim == 2
 
+        condition, _ = pack([condition], 'b * d') # handle condition with ndim 2 - (batch, dim)
+
         if one_headed:
             codes = rearrange(codes, 'c d -> 1 c d')